Processing point clouds of vehicle sensors having variable scan line distributions using voxel grids

1. A method for processing point clouds having variable spatial distributions of scan lines, the method comprising: receiving a point cloud portion corresponding to an object in an environment through which a vehicle is moving, the point cloud portion including a plurality of scan lines arranged according to a particular spatial distribution; constructing, by one or more processors, a voxel grid corresponding to the received point cloud portion, wherein the voxel grid includes a plurality of volumes in a stacked, three-dimensional arrangement, and constructing the voxel grid includes (i) determining an initial classification of the object, (ii) setting one or more parameters of the voxel grid based on the initial classification, and (iii) associating each volume of the plurality of volumes with an attribute specifying how many points, from the point cloud portion, fall within that volume; and generating, by one or more processors and using the constructed voxel grid, signals descriptive of a current state of the environment through which the vehicle is moving.

2. The method of claim 1 , wherein setting one or more parameters of the voxel grid based on the initial classification includes setting one or more dimensions of a leaf size of the voxel grid, the leaf size defining a real-world volume corresponding to each of the plurality of volumes.

3. The method of claim 2 , wherein setting the one or more dimensions of the leaf size includes using the initial classification to access a database storing data that associates different leaf sizes with different object classifications.

4. The method of claim 1 , wherein setting one or more parameters of the voxel grid is based on (i) the initial classification and (ii) an indication of the particular spatial distribution.

5. The method of claim 4 , wherein setting the one or more dimensions of the leaf size includes using the initial classification and the indication of the particular spatial distribution to access a database storing data that associates different leaf sizes with different pairs of object classifications and spatial distributions.

6. The method of claim 1 , wherein the particular spatial distribution of scan lines comprises a uniform distribution, a Gaussian distribution, a multimodal distribution, and/or an arbitrary distribution.

7. The method of claim 1 , further comprising: partitioning a point cloud frame into a plurality of portions in accordance with probable boundaries between separate physical objects, wherein the point cloud portion is one of the plurality of portions.

8. The method of claim 7 , wherein generating the signals descriptive of the current state of the environment through which the vehicle is moving includes: determining, based on the constructed voxel grid, a final classification of the object.

9. The method of claim 8 , wherein the initial classification is a general class and the final classification is a specific class within the general class.

10. The method of claim 8 , wherein the initial classification is an initial prediction of an object class and the final classification confirms or refutes the initial prediction.

11. A non-transitory computer-readable medium storing thereon instructions executable by one or more processors to implement a self-driving control architecture of a vehicle, the self-driving control architecture comprising: a perception component configured to receive a point cloud frame generated by a sensor configured to sense an environment through which the vehicle is moving, the point cloud frame including a plurality of scan lines arranged according to a particular spatial distribution, partition the point cloud frame into a plurality of portions in accordance with probable boundaries between separate physical objects, each of the plurality of portions corresponding to a respective one of a plurality of objects, construct a voxel grid corresponding to a first portion of the plurality of portions and a first object of the plurality of objects, wherein the voxel grid includes a plurality of volumes in a stacked, three-dimensional arrangement, and constructing the voxel grid includes (i) determining an initial classification of the first object, (ii) setting one or more parameters of the voxel grid based on the initial classification, and (iii) associating each volume of the plurality of volumes with an attribute specifying how many points, from the first portion, fall within that volume, and generate, using the constructed voxel grid, signals descriptive of a current state of the environment through which the vehicle is moving; and a motion planner configured to generate driving decisions based on the signals descriptive of the current state of the environment, and cause one or more operational subsystems of the vehicle to maneuver the vehicle in accordance with the generated driving decisions.

12. The non-transitory computer-readable medium of claim 11 , wherein the one or more parameters of the voxel grid include one or more dimensions of a leaf size of the voxel grid, the leaf size defining a real-world volume corresponding to each of the plurality of volumes.

13. The non-transitory computer-readable medium of claim 12 , wherein the perception component is configured to set the one or more dimensions of the leaf size at least by using the initial classification to access a database storing data that associates different leaf sizes with different object classifications.

14. The non-transitory computer-readable medium of claim 11 , wherein the perception component is configured to set the one or more parameters of the voxel grid based on (i) the initial classification and (ii) an indication of the particular spatial distribution.

15. The non-transitory computer-readable medium of claim 11 , wherein the perception component is configured to set the one or more dimensions of the leaf size at least by using the initial classification and the indication of the particular spatial distribution to access a database storing data that associates different leaf sizes with different pairs of object classifications and spatial distributions.

16. The non-transitory computer-readable medium of claim 11 , wherein the perception component is further configured to: partition a point cloud frame into a plurality of portions in accordance with probable boundaries between separate physical objects, wherein the point cloud portion is one of the plurality of portions.

17. The non-transitory computer-readable medium of claim 16 , wherein the perception component is configured to generate the signals descriptive of the current state of the environment through which the vehicle is moving at least by: determining, based on the constructed voxel grid, a final classification of the object.

18. The non-transitory computer-readable medium of claim 17 , wherein the initial classification is a general class and the final classification is a specific class within the general class.

19. The non-transitory computer-readable medium of claim 17 , wherein the initial classification is an initial prediction of an object class and the final classification confirms or refutes the initial prediction.

20. A vehicle comprising: a sensor configured to sense an environment through which the vehicle is moving and generate point cloud frames; one or more operational subsystems; and a computing system configured to receive a point cloud frame generated by the sensor, the point cloud frame including a plurality of scan lines arranged according to a particular spatial distribution, partition the point cloud frame into a plurality of portions in accordance with probable boundaries between separate physical objects, each of the plurality of portions corresponding to a respective one of a plurality of objects, construct a voxel grid corresponding to a first portion of the plurality of portions and a first object of the plurality of objects, wherein the voxel grid includes a plurality of volumes in a stacked, three-dimensional arrangement, and constructing the voxel grid includes (i) determining an initial classification of the first object, (ii) setting one or more parameters of the voxel grid based on the initial classification, and (iii) associating each volume of the plurality of volumes with an attribute specifying how many points, from the first portion, fall within that volume, generate, using the constructed voxel grid, signals descriptive of a current state of the environment through which the vehicle is moving, generate driving decisions based on the signals descriptive of the current state of the environment, and cause the one or more operational subsystems of the vehicle to maneuver the vehicle in accordance with the generated driving decisions.

21. The vehicle of claim 20 , wherein the one or more parameters of the voxel grid include one or more dimensions of a leaf size of the voxel grid, the leaf size defining a real-world volume corresponding to each of the plurality of volumes.